Claim Verification: "Training and running today's frontier AI models consumes more electricity than entire small countries." — Proved

Automated fact-verification of the claim: "Training and running today's frontier AI models consumes more electricity than entire small countries." Verdict: PROVED Key Findings A typical AI-focused data centre (which the International Energy Agency (IEA) defines as a facility performing both AI training and inference) consumes 1,050 GWh of electricity per year — equivalent to 100,000 US households. Nauru, the smallest UN member state with documented electricity data (~11,000 people), consumes 37.89 GWh per year — the entire nation. A single typical AI-focused data centre therefore consumes 27.7 times more electricity annually than all of Nauru (1,050 GWh vs. 37.89 GWh). Independent cross-check: a peer-reviewed 2025 study (Harding & Moreno-Cruz, Environmental Research Letters) found US AI electricity alone is comparable to Iceland's total electricity (~19,580 GWh) — 517 times more than Nauru. Files proof.py — Re-runnable Python verification script proof.md — Structured proof report proof_audit.md — Full verification audit trail proof_narrative.md — Plain-language summary proof.json — Machine-readable structured data Generated by Proof Engine v0.10.0.针对声明“当前前沿人工智能模型的训练与运行能耗超过小型主权国家全年总耗电量”的自动化事实核查 裁决：已证实（PROVED） 关键发现 国际能源署（International Energy Agency, IEA）将同时开展人工智能训练与推理的设施定义为人工智能专用数据中心，单座此类典型数据中心的年耗电量可达1050吉瓦时（GWh），相当于10万户美国家庭的年用电总量。 瑙鲁是有官方用电记录的最小联合国会员国（人口约1.1万），其全国全年总耗电量仅为37.89吉瓦时（GWh）。 因此，单座典型人工智能专用数据中心的年耗电量是瑙鲁全国总耗电量的27.7倍（1050吉瓦时对比37.89吉瓦时）。 独立交叉验证：2025年的同行评议研究（Harding与Moreno-Cruz，《Environmental Research Letters》）显示，仅美国国内的人工智能用电规模就可与冰岛全年总耗电量（约19580吉瓦时）相当——该数值是瑙鲁全国耗电量的517倍。 文件 proof.py —— 可重复运行的Python验证脚本 proof.md —— 结构化核查报告 proof_audit.md —— 完整核查审计轨迹 proof_narrative.md —— 通俗语言总结报告 proof.json —— 机器可读结构化数据 本报告由Proof Engine v0.10.0生成。